Diagnostic spectral features for plastic litter detection in freshwater environments

Developing spectral-based algorithms for detecting plastics in natural environments requires an understanding of the spectral properties of plastic materials. However, spectral characterization of plastics and identification of diagnostic spectral features for plastic detection in natural freshwater environments remain unexplored. Unlike earlier studies that identified diagnostic spectral absorption features of plastics based on a single dominant background (either water-only or sand-only), we sought to identify features that account for the coexistence of multiple dominant background materials within the field of view. Spectral measurements were performed on plastic samples collected from the shores of Lake Victoria in Uganda using an SR-3501 Spectroradiometer. Three background surfaces (water, sand and vegetation) representative of the natural environment were selected and used as backgrounds for conducting spectral measurements. Several absorption features in plastic spectra, which aligned with earlier studies, were detected. However, the strength of these absorption features varied significantly across background surfaces, with coefficients of variation exceeding 20% for most features. Diagnostic spectral features for plastics were therefore identified based on freshwater shore zones. Absorption features centered at 1086nm, 1660nm, 1681nm, 1714nm, 1729nm, 2054nm, 2132nm and 2156nm were identified as diagnostic of plastic materials. Particularly, the 1729nm feature was detected in more plastic samples than any other, making it particularly useful for developing plastic detection algorithms. While previous studies identified plastic absorption features in the ⁓930nm-978nm and ⁓1140nm-1220nm spectral ranges, this study revealed that, within these ranges, plastics and backgrounds have similar absorption characteristics and therefore cannot be spectrally differentiated using these absorption features.